Enhancement: Add project scaffolding for docs/solutions/ schema + agentic feedback loops

[revivalfightwear]

Enhancement: Add project scaffolding for docs/solutions/ schema + agentic feedback loops

---

Labels

enhancement, documentation

---

Body

## Summary
 
First-time use of `/workflows:compound` on a Shopify/Node.js project revealed a bootstrapping gap that undermines the knowledge compounding system.
 
**The core issue**: The plugin has a well-defined YAML schema (`skills/compound-docs/references/yaml-schema.md`) with specific enum values for `problem_type`, `component`, `root_cause`, etc. But this schema is never communicated to new projects, so:
 
1. First compound run may create docs with different/inconsistent frontmatter
2. Future `learnings-researcher` queries can't find them (expects exact enum matches)
3. Knowledge doesn't actually compound
 
## What Happened
 
When `/workflows:compound` ran for the first time, the agents created docs in `docs/solutions/` with YAML frontmatter. But the receiving project had:
 
- No `docs/solutions/README.md` explaining the expected structure
- No knowledge of the schema the agents expect
- No way to validate the frontmatter is correct
- No guarantee future compound runs will find past solutions
 
## Suggested Improvements
 
### 1. Add `/compound:init` command (Short-term)
 
A one-time setup that:
- Creates `docs/solutions/README.md` with the schema definition
- Creates category subdirectories matching `problem_type` enums
- Optionally creates `docs/solutions/.schema.yaml` for validation
 
### 2. Make `/workflows:compound` check for scaffolding (Medium-term)
 
Before creating documentation:
- If `docs/solutions/README.md` doesn't exist, auto-scaffold or prompt user
- Validate generated frontmatter against schema before writing
 
### 3. Add agentic feedback loops (Long-term)
 
**The meta-problem**: When an agent discovers issues while using the plugin, there's no standardized way to submit feedback agentically.
 
Suggestion: Add `/compound:feedback` command that:
- Captures context automatically (plugin version, command used, what happened)
- Formats feedback according to maintainer preferences
- Submits via GitHub issue
 
This creates a virtuous cycle:

Agent uses plugin → Discovers issue → Submits feedback → Plugin improves → Better agent experience

 
**This issue itself is an example** - an agent (Claude) discovered these gaps but had no agentic way to submit them. The user had to manually identify the correct repo.
 
## Bug Found
 
The installed plugin's `/report-bug` command (v2.27.0) targets:

kieranklaassen/every-marketplace

 
But the actual public repo is:

EveryInc/compound-engineering-plugin

 
## Related Issues
 
- #95 (pre-compact knowledge capture) - shares the "auto-capture before context loss" philosophy
- #79 (docs discoverability) - same theme of hidden resources
 
## Environment
 
- **Plugin Version**: 2.27.0
- **Project Type**: Shopify Liquid + Node.js
- **First-time compound user**: Yes
 
## Impact
 
Without this fix:
- First-time users create docs with inconsistent schemas
- Knowledge doesn't compound across runs
- `learnings-researcher` can't effectively find past solutions
- The core value prop ("each unit of work makes future work easier") is undermined
 
---
 
*Feedback discovered and documented by Claude agent during first `/workflows:compound` use*

---